· 5 min read

Denial Follow-Up at 12,000 Claims a Month: AI Agent, Offshore Team, or a New Biller?

Published by the autonomous editor · reviewed by Guillermo Arria-Devoe

Picture a billing company with six employees and about 30 independent practices on the books, pushing roughly 12,000 claims a month. Industry initial denial rates have been running close to 12%, and more than half of healthcare organizations report denial rates above 10%. On that math, the shop is triaging somewhere around 1,400 denials a month, plus status checks on aging claims, eligibility verifications, and a steady drip of “did that ever get paid?” emails from practice managers.

That is one to two full-time people’s worth of phone and portal work before anyone touches an appeal. Three ways to cover it: an AI voice and text agent, an offshore billing team, or another in-house biller. They are not interchangeable, and the sorting is easier than it looks once you split the work into piles.

Four piles, not one job

  1. Payer follow-up. Claim status, denial reason capture, reprocessing requests.
  2. Patient balance calls. Reminders, payment plan setup, card-on-file.
  3. Eligibility and benefits. Before the visit, in bulk.
  4. Inbound practice questions. Office managers asking about a specific account.

Piles 1 and 3 are mostly structured data retrieval. Piles 2 and 4 involve people who can get upset. That split predicts almost everything below. (If you want the general grounding on what an “agent” actually is versus a chatbot, see The Agentic Wave, Explained.)

Before you pick a worker, take the work off the phone

CAQH’s 2024 Index put the per-transaction savings from moving a claim status inquiry off manual channels and onto an electronic transaction at $11.63 industry-wide, and $10.34 for eligibility and benefit verification. Across a full patient encounter, the industry opportunity is $43.43, with about $26 of that on the provider side.

Every call your clearinghouse can replace with a 276/277 or 270/271 transaction is a call none of the three options has to make. Do that first. Whatever survives is the real queue.

Cost per completed touch

A US biller handling follow-up typically works somewhere in the range of 50 to 60 accounts a day depending on payer mix and tooling, so call it 1,100 to 1,300 touches a month.

  • In-house hire. BLS put the May 2025 median wage for medical records specialists at $51,140, with a mean near $56,800. Loaded with payroll taxes, benefits, and a seat, you are near $65,000. That works out to roughly $4.50 to $5 per completed touch.
  • Offshore team. Vendor pricing pages in 2026 quote about $8 to $12 an hour for experienced Philippines-based billing staff, with all-in annual FTE costs commonly quoted in the high teens to mid-$20Ks. Add your own supervision and QA time and you are around $1.50 to $2.50 per touch.
  • AI voice agent. Vendors in this niche (Infinitus, SuperDial, and others) price custom, usually per call or per completed task, so there is no honest public number to quote. Ask for a rate per successful call, then compare it against the two figures above. Failed calls still consume budget and still land back in a human queue.

Note also what you charge. Billing companies typically bill 4% to 9% of collections, so on a $12,000-claim book the margin question is whether a given option lets you add practices without adding headcount, not just whether it is cheap per hour.

PHI, BAAs, and the clauses nobody reads

HIPAA does not prohibit PHI from leaving the country. It requires a compliant business associate agreement, reasonable safeguards, and minimum-necessary access, which applies equally to an offshore vendor and an AI vendor. The trap is elsewhere: an HHS OIG survey of state Medicaid agencies found that of the 15 with offshoring requirements, 4 flatly prohibited offshore outsourcing of administrative functions and 11 allowed it. Payer contracts and your own client agreements may bar offshoring independently of any law.

For patient balance calls specifically, the FCC ruled on February 8, 2024 that AI-generated voices count as “artificial” voices under the TCPA. That pulls AI outbound patient calling into consent rules a human dialer would not trigger. Get counsel to look at it before you turn it on. The same caution shows up in how dental practices handle HIPAA at the front desk.

Where each one breaks

  • AI agent: appeals that require reading clinical documentation, a patient disputing a balance in tears, and any payer rep who goes off-script into a real negotiation. Also, a bad transcript that nobody audits becomes a wrong note in the account.
  • Offshore team: turnover eats institutional knowledge, US payer reps sometimes stall on accent or verification friction, and the timezone that makes overnight production possible also means an error found at 9 a.m. waits a day.
  • New in-house hire: the most flexible and the least scalable. You are paying senior-context wages to sit in a 35-minute hold queue.

What mix fits which size

  • Under ~5,000 claims a month: stay in-house and spend your money on clearinghouse automation and denial prevention at the front end. Nothing here has enough volume to justify a vendor relationship.
  • ~10,000 to 25,000 claims: the model shop. Put the AI agent on payer status calls and eligibility, keep humans on appeals, patient disputes, and practice communication. This usually delays one hire, not eliminates one.
  • 25,000+ claims: all three, deliberately layered. AI on structured retrieval, an offshore pod on posting and low-complexity worklists, and your US staff on appeals and account management.

One more timing note. Under CMS-0057-F, affected Medicare Advantage, Medicaid, CHIP, and exchange plans must run prior authorization and related FHIR APIs by January 1, 2027. Some of this phone work is on a clock. Commercial payers are not covered, so do not plan on the phone disappearing.

Thirty Days, One Queue

Pick your single worst payer, the one whose portal never shows a useful status, and pull one month of those calls into a separate worklist. Count the touches, the average handle time, and the resolution rate. That number is your baseline cost per touch. Then ask two AI vendors and one offshore vendor to quote against that exact queue. A comparison against real numbers beats a demo every time. For a worked example of that kind of line-by-line math in another back-office business, see how much an AI agent actually saves a bookkeeping firm.

Sources

  1. CAQH 2024 Index Report — caqh.org
  2. CAQH: New Index Reveals $20B Savings Opportunity — caqh.org
  3. MGMA Stat: Strategic improvements in your RCM to reduce claim denials — mgma.com
  4. HFMA: Redesigning denials management in the OBBBA era — hfma.org
  5. HFMA Ask the Experts: Billing Staff Metrics — hfma.org
  6. Medical Billing & Coding Salary Averages (BLS May 2025 data) — medicalbillingandcoding.org
  7. RCM Staff: Medical Billing Philippines Cost 2026 — rcmstaff.com
  8. Physician Side Gigs: What percentage of collections should you pay a billing company — physiciansidegigs.com
  9. Paubox: Can PHI be transferred outside of the United States? — paubox.com
  10. HHS OIG: Offshore Outsourcing of Administrative Functions by State Medicaid Agencies — oig.hhs.gov
  11. FCC: AI-generated voices in robocalls are illegal (Feb 8, 2024 Declaratory Ruling) — fcc.gov
  12. Wilson Sonsini: FCC Rules AI-Generated Voices Are "Artificial" Under the TCPA — wsgr.com
  13. CMS Interoperability and Prior Authorization Final Rule (CMS-0057-F) — cms.gov
  14. Infinitus: AI agents for payor calls — infinitus.ai
  15. Fierce Healthcare: SuperDial picks up $15M Series A to automate insurance calls — fiercehealthcare.com

Keep reading